Ya. I understand VPN. I do enterprise IT stuff. The things I build assume a secure environment. VPN is step one.
Nailing down a web server on the internet tho ... there's so many ways to attack. There's so many things to secure. And its a bit complex to manage all that.
The nextcloud site covers hardening the server, but doesn't even mention vpn.
I've been watching threads like this. I'm pretty convinced vpn is the answer.

A honeypot is something that is intentionally left available, to alert you when it gets hit. In practice, they’re just a tool to tell security specialists when they need to start worrying; They wouldn’t be used by the average user at all.

The goal is to build your security like layers, and ideally have all of your services behind the secure walls. Between these layers, you have honeypots. If someone gets through your first layer of security but hits the honeypot, you know someone is sniffing around, or maybe has an exploit for your outer layer that you need to research. If they get through the second layer and hit your second honeypot, you know that someone is specifically targeting you (instead of simply running automated scans) and you need to pay closer attention. Etc…

Reinforcing the attack layer comes in two main forms, which work in tandem: Strengthening the actual layer, and reducing attack vectors. The first is focused on using strong passwords, keeping systems up to date, running something like Fail2Ban for services that are exposed, etc… The goal is for each layer of security to be robust, to reduce the chances of a bot attack actually working. Bots will simply sniff around and automatically throw shit at the wall to see if anything sticks.

The second part is focused on identifying and mitigating attack vectors. Essentially reducing the amount of holes in the wall. It doesn’t matter how strong the wall is if it’s full of holes for your server’s various services. The goal is typically to have each layer be as solid as possible, and grant access to the layers below it. So for instance, running a VPN. The VPN gets you access to the network, without exposing services externally. In order to access your services, they need to get through the VPN first, making the VPN the primary attack vector. So you can focus on ensuring that the VPN is secure, instead of trying to spread your focus amongst a dozen different services. If it’s exposed to the open internet, it is a new potential attack vector; The strength of the wall doesn’t actually matter, if one of those services has an exploit that someone can use to get inside your network.

Home users really only need to worry about things like compromised services, but corporate security specialists also focus on things like someone talking their way past the receptionist and into the server room, USB sticks getting “lost” around the building and plugged into random machines by curious employees, etc… All of these are attack vectors, even if they’re not digital. If you have three or four layers of security in a corporate setting and your third or fourth honeypot gets hit, you potentially have some corporate spy wrist-deep in your server room.

For an easy example, imagine having a default password on a service, and then exposing it to the internet via port forwarding. It doesn’t matter how strong your firewall is anymore. The bot will simply sniff the service’s port, try the default credentials, and now it has control of that service.

The better way to do it would be to reduce your attack vectors at each layer; Require the VPN to access the network via a secure connection, then have a strong password on the service so it can’t easily be compromised.